Princeton Researcher Uses AI to Match Georgia Voters to Their Ballots
Max Springer, a postdoctoral research fellow at Princeton University's Center for Information Technology Policy, used an AI coding agent and public records to reconstruct the casting order of 1.52 million ballots from Georgia's May 2026 primary, as he detailed in a university blog post. That is 98.9% of in person ballots in the 114 of 139 counties where he found the order could be recovered.
Springer did not discover the flaw. He pointed the agent at an October 2022 research paper showing that the scanners' supposedly random ballot numbering is deterministic and can be reversed, then gave it two public files: each county's early voting list and the cast vote record. He wrote that within a couple of hours he had a working pipeline, and that at no point did the agent refuse or raise concerns.
Those two files alone uniquely tied about 1% of analysed early in person voters to a ballot. Adding each location's public scanner audit log and check in records went much further. In Heard County, the agent matched the majority of 650 early in person voters to a specific ballot and the rest to within a single swap, and the result held across all 1,860 early voters at the Ball Ground vote center in Cherokee County.
A corrected software version was certified in March 2023. Springer recommends that jurisdictions using Dominion scanners install version 5.17 or newer, and argues that AI coding tools have removed the expertise that once kept vulnerabilities like this one from being exploited.
The Associated Press reported on 13 September that Dominion has since been sold and renamed Liberty Vote, and that Georgia is the only state using the system statewide. According to the AP, the secretary of state's office will black out problematic information in records released to the public and is working with a vendor to scramble the original order, while the State Election Board rejected a proposal that would have required the office to apply the manufacturer's recommended update, with some members citing the short timeline before the November midterms. The AP noted that the flaw does not allow votes to be changed.
We hope you enjoyed this article
Consider subscribing to one of our newsletters like Cybersecurity AI Weekly or Daily AI Brief.
Also, consider following us on social media:
More from Cybersecurity
Sep 13 Anthropic Blocks Yemen Cell Using Claude Code for Missile Software Sep 13 Researchers Link OpenAI Agents to May RubyGems Attack Sep 11 Anthropic Publishes Five Cases of Claude Use That Could Support Biological Weapons Work Sep 11 Senate Opens Inquiry Into OpenAI Agents' Hugging Face Hack Sep 11 FAZE Security Emerges From Stealth With $6 Million Seed RoundCybersecurity AI Weekly
Weekly newsletter about AI in Cybersecurity.
Market report
2025 Generative AI in Professional Services Report
Thomson Reuters
This report by Thomson Reuters explores the integration and impact of generative AI technologies, such as ChatGPT and Microsoft Copilot, within the professional services sector. It highlights the growing adoption of GenAI tools across industries like legal, tax, accounting, and government, and discusses the challenges and opportunities these technologies present. The report also examines professionals' perceptions of GenAI and the need for strategic integration to maximize its value.
Read moreYou may also like
Quebec Parties Join AI Conduct Code Before October Election
US Agencies Accuse Chinese AI Firms of Industrial Scale Model Distillation
Researchers Link OpenAI Agents to May RubyGems Attack
Anthropic's Threat Report Finds AI Moving From Assistant to Orchestrator
Anthropic Assesses Four Incidents Where Claude Models Reached the Real Internet
Daily AI Brief: the AI news that matters, in your inbox.